pi-provider-vancine

Use Vancine from Pi through /login and a dynamic /model list, without writing ~/.pi/agent/models.json.

Install

pi install npm:pi-provider-vancine

Then start Pi.

Log in

1. In Pi, run /login.
2. Choose Vancine.
3. Paste your Vancine API key into the secret prompt.
4. Pi stores the key in its credential store. This extension does not write its own credential file.

Choose a model

1. Run /model.
2. Select a model under the Vancine provider.
3. You do not need to create or edit models.json for this basic path.

/model loads Vancine's current compatible Chat Completions models from the production catalog at https://vancine.com/api/pi/catalog. The list is not a hardcoded whitelist.

Catalog refresh and fallback

The live source is GET https://vancine.com/api/pi/catalog. See docs/catalog-endpoint.md.

• While Pi has this extension loaded, the provider refreshes the catalog at most every 4 hours, and sends If-None-Match / If-Modified-Since when it already has validators.
• Pi 0.85.1's standalone pi update --models creates a ModelRuntime from built-in providers only. It does not load third-party extensions, so it cannot refresh Vancine immediately. This package does not add a separate refresh command and does not claim that Pi upstream can force-refresh third-party catalogs on its own.
• A cached deepseek-flash entry is treated as retired. On the next catalog refresh while this extension is loaded, the provider bypasses the four-hour window, fetches the current catalog without the old validators, and drops that ID. It does not alias or rewrite the old ID. Do not edit or delete ~/.pi/agent/models-store.json by hand.
• When a catalog request succeeds, /model shows the compatible Chat Completions models from that response, including a legitimate empty list.
• If a later catalog request fails, Pi keeps the last successful cached catalog, including a cached empty list. This extension does not overwrite a valid cache with fallback models.
• The four-model fallback is used only when there is no cache at all, network refresh is allowed, and that first catalog request fails:
  • hy4-preview
  • deepseek-v4.1-flash
  • glm-5.3-flash
  • qwen3.8-flash
• Offline startup without a cache shows no Vancine models until a network refresh is attempted.
• After a successful catalog refresh, delisted models disappear and are not revived by fallback.

The catalog is not guaranteed to be realtime and can fail. Fallback models are a static first-run snapshot, not live model sync.

Vancine API

Base URL	https://vancine.com/v1
Protocol	OpenAI-compatible Chat Completions
Register	https://vancine.com
API keys	https://vancine.com/console
Models and pricing	https://vancine.com/docs/models
Public pricing JSON	https://vancine.com/api/pricing
Agent notes	https://vancine.com/docs/agents

Security

• The API key is stored by Pi's credential store after /login.
• This extension does not keep its own copy of the key, does not put the key in the model cache, and does not send the key to catalog, npm, GitHub, Models.dev, or other third-party hosts.
• Authorization is only used for Vancine API requests that Pi itself sends to https://vancine.com.
• This package has no telemetry, analytics, install scripts, or automatic edits to your Pi settings.

Troubleshooting

Provider does not appear in /login

• Confirm the package is installed: pi list.
• Restart Pi after install.
• Check that the extension was not disabled in pi config.

No models after /login

• Run /model after login. Models are listed only when Vancine auth is configured.
• If this is a first run, network refresh is allowed, and the catalog request fails, you should see the four offline fallback models.
• If a previous successful refresh stored an empty compatible catalog, /model stays empty.

Model catalog is temporarily unavailable

• Pi keeps the last successful catalog when one exists, including an empty catalog.
• First run without a cache, after a failed network refresh, uses the four offline snapshot models. That is not live sync.
• The catalog can fail; it is not claimed to be always live.

Model list still shows a retired ID, or a renamed model is missing

• deepseek-flash is a retired Vancine ID. This provider migrates a cache that still contains it the next time Pi loads the extension and runs a catalog refresh: it fetches the live catalog without the old ETag, then keeps the server list. It does not map the old ID to a new one.
• Pi 0.85.1 pi update --models does not load this extension, so it will not perform that migration. Starting Pi (or otherwise loading the extension so it can refresh) is what applies the cache migration.
• Do not hand-edit or delete ~/.pi/agent/models-store.json.
• Updating the npm package does not by itself invent a model. If production /api/pi/catalog does not list the model, a package update cannot add it.

API key is invalid

• Create or copy a key from the Vancine console.
• Run /login again and choose Vancine.
• This extension never prints the full key in errors.

Compatibility errors

• Vancine Chat Completions is OpenAI-compatible. This provider sets supportsDeveloperRole: false so Pi sends system instead of developer.
• supportsReasoningEffort is not force-disabled. In the offline fallback only deepseek-v4.1-flash declares it as true (a verified Vancine Chat Completions fact); every other model leaves it unset so Pi's own default still applies. The live catalog carries the same per-model value.
• If a specific model rejects a reasoning parameter, report it with the model id and error text (redact the key).

Update

pi update npm:pi-provider-vancine

Remove

pi remove npm:pi-provider-vancine

Independence

Vancine is an independent OpenAI-compatible API service.

This package is a community extension published and maintained by Vancine. It is not a built-in Pi provider.

It is not an official extension, partnership, certification, or endorsement by Pi, Earendil Works, or their maintainers.

Listing on Models.dev or any other directory is not an official partnership, endorsement, or approval by Pi or Earendil Works.

This package does not claim that Vancine is built into Pi, that Pi has officially endorsed it, or that every dynamic model was paid-tested in this release.